feat(shortcuts): recordable recently-used-tabs gesture and a Search everywhere chord

[AppsOfUse]

Summary

Recently used tabs and Search everywhere were fixed gestures with only a mode select.
Both are now the user's to set in Settings > Shortcuts.

Recently used tabs records a hold gesture: one modifier to hold (Ctrl, Option/Alt, or Cmd on macOS) and a key to tap, stored in the new synced pref recentTabsGesture.
The tap key is recorded by KeyboardEvent.code, because Option turns ` into a dead key and letters into other characters, so key would never match its own next press.
Absent means Ctrl+Tab, so there is no migration.
The recorder refuses, and says why, naming keys that do work with that modifier:
two or no modifiers; the OS's own chords (Cmd+Tab, Cmd+Space, Cmd+, Ctrl+Space; Alt+Tab, Alt+F4, Alt+ and Win/Super elsewhere);
Ctrl with a letter, digit, or a key that makes a control character ([ \ ] / -, readline's and the terminal's);
Cmd with a letter or digit (copy, paste, quit...);
Option with a letter or digit on macOS (it types), and Alt with any character key elsewhere (Meta to the terminal);
arrows, Home, End and Page keys; Enter, Space and Backspace.
Option with punctuation is accepted on macOS, matching the chord recorder; Option+then takes the à/è dead key while the gesture is on, whichdocs/shortcuts.md` states as the user's trade-off.
A synced gesture this platform cannot use falls back to Ctrl+Tab without rewriting the stored value, and the row says so with a reset.

Search everywhere becomes a normal shortcut entry with a chord that ships unbound, and keeps its double-Shift select on the same row.
It is in PASS_TO_APP, so on macOS the chord works from inside a terminal (elsewhere only with Shift or Alt, as before), and the command palette shows it.

Conflicts move into a pure conflicts(): chord vs chord as before, plus gesture vs chord while the gesture is on, flagged on both rows.
The gesture's mode select now reads Off / On, since the keys sit on the recorder.

Docs: docs/shortcuts.md.

Review follow-ups

A review found defects in the new code; all are fixed here, each with a unit test that failed before its fix:

• On macOS a Cmd chord in PASS_TO_APP also swallowed the same Ctrl key in a terminal (bind Cmd+R, and Ctrl+R was dead). terminalPassesToApp now requires Cmd on macOS. This also fixes Ctrl+P in a terminal, which was dead on main because of Cmd+P.
• The refusals above (Cmd+letter would have killed copy app-wide; control-character and cursor keys; Alt+F4).
• An auto-repeat as the first keydown of a walk left the gesture pinned and recent-places recording suspended.
• A gesture also stores the character it typed, so it displays and conflicts correctly on non-US layouts.
• conflicts() is platform-aware (on macOS only a Cmd hold conflicts with a Cmd chord); the logo key is Super on Linux.
• Docs and comments corrected, docs/shortcuts.md wrapped.

Testing

Manually tested by the author in the dev app: rebinding Recently used tabs and recording a Search everywhere chord. Both behaved as expected.
That pass was before the review follow-ups above, which have not been exercised by hand yet.

• npm test (3233 passed), npx tsc -b, npm run typecheck:e2e
• e2e: settings.e2e.ts 123 passing on three consecutive runs, codenav.e2e.ts 31, tabs-layout.e2e.ts 45, app.e2e.ts 40.
• Controls run on macOS: removing search-everywhere from PASS_TO_APP turns "opens Search Everywhere on a recorded chord, from a focused terminal" red; restoring the old predicate in terminalPassesToApp turns "leaves ⌃R to the shell when ⌘R is bound" red. Both green again with the code as committed.
• New: "sends nothing to the PTY while it walks on an Option hold" (tabs-layout.e2e.ts).

Pre-existing on main, found during review, not addressed here

• A corrupt shortcutBindings blob falls into the outer catch {} of loadShortcuts, and the next save wipes every customisation.
• The hint lines for Previous tab, Next tab and New tab show the raw key (shortcuts.defs.tabPrev.hint). shortcutHint compares against the key with its settings: prefix, which i18next does not return on a miss, so a missing hint is never detected (since feat(i18n): the shortcut rows speak Chinese #391).

Note: #397 also edits tabs-layout.e2e.ts; whichever merges second needs the fixedWaits.test.ts count rechecked.

🤖 Generated with Claude Code

[simion]

Thanks, nice addition. Appreciate you jumping in with two PRs already 🧡

[simion]

Approved, but this now has merge conflicts (likely from #400 landing). Needs a rebase before merge.